Cast pack for a series Format run: the 30-file attachment budget

One Format run carries 30 media files in total, at most 30 images, 10 videos and 10 audio, counting URLs in input. How to fit a cast and two locations in.

4 min readSume
All posts

A single Format run can see 30 media files in total: attachments take up to 30 images, and HTTPS media URLs found anywhere inside input count against the same budget of 30, with at most 30 images, 10 videos and 10 audio files. That is enough for a series cast pack of a few characters, a couple of locations and a handful of voice samples, if you plan it. Go over and the create call fails with 400 invalid_attachment.

The budget

Media budget of one Format run (read 2026-10-07)
RuleLimit
Total media files (attachments plus URLs in input)30
ImagesAt most 30
VideosAt most 10
Audio filesAt most 10
input size2 MiB, at most 64 top-level keys
instruction carried to the runAbout the first 4,000 characters of the 8,000 accepted
Error when exceeded400 invalid_attachment

A cast pack that fits

Plan the 30 like a casting budget. Four principal characters at three views each (front, three-quarter, profile) take 12 images. Two locations at three angles each take 6 more. A wardrobe or prop sheet per character takes 4. That is 22 images, leaving 8 spare, plus up to 10 videos for motion references and up to 10 audio files for voice samples, which brings the file count against the same total of 30 rather than on top of it, so count them together.

Keep it in input, not in the prompt

Sume writes input to a file in the run's workspace and tells the agent it is data from the caller, not instructions. Put the cast list there as an object with a name, a role and the image URLs for each character, and keep the instruction short: 'Render episode 9 with the cast in the input.' Nested keys do not count towards the 64-key limit, so a cast object with many characters is one key.

Use attachments for images you want the agent to look at directly, and input for the structured facts that go with them. Media URLs in either place are counted once against the budget.

When the cast is bigger than the budget

Split by scene rather than trimming the cast. A scene with two characters needs two characters' views, not twelve people's. Build each episode's run from the characters in its scene list and send only those, with the location for that scene. A bulk queue of up to 100 items makes the per-item packs practical, since each item is its own Format-run body with its own attachments.

Every URL must be public HTTPS. An item with a dead URL fails at create, before any queue exists, so check your pack once with a pilot run before you queue the season.

Sources

Related posts

More in Formats

All Formats posts

Written by Sume